[1]胡丹丹,张忠婷.基于改进YOLOv5s的面向自动驾驶场景的道路目标检测算法[J].智能系统学报,2024,19(3):653-660.[doi:10.11992/tis.202206034]
HU Dandan,ZHANG Zhongting.Road target detection algorithm for autonomous driving scenarios based on improved YOLOv5s[J].CAAI Transactions on Intelligent Systems,2024,19(3):653-660.[doi:10.11992/tis.202206034]
点击复制
HU Dandan,ZHANG Zhongting.Road target detection algorithm for autonomous driving scenarios based on improved YOLOv5s[J].CAAI Transactions on Intelligent Systems,2024,19(3):653-660.[doi:10.11992/tis.202206034]
基于改进YOLOv5s的面向自动驾驶场景的道路目标检测算法
《智能系统学报》[ISSN 1673-4785/CN 23-1538/TP] 卷:
19
期数:
2024年第3期
页码:
653-660
栏目:
学术论文—智能系统
出版日期:
2024-05-05
- Title:
- Road target detection algorithm for autonomous driving scenarios based on improved YOLOv5s
- Keywords:
- YOLOv5s; autonomous driving; target detection algorithm; depthwise separable convolution; receptive field block; adaptive spatial feature fusion; PANet; multiscale feature fusion
- 分类号:
- TP391.4
- 文献标志码:
- 2023-09-14
- 摘要:
- 在复杂道路场景中检测车辆、行人、自行车等目标时，存在因多尺度目标及部分遮挡易造成漏检及误检等情况，提出一种基于改进YOLOv5s的面向自动驾驶场景的道路目标检测算法。首先，利用深度可分离卷积替换部分普通卷积，减少模型的参数量以提升检测速度。其次，在特征融合网络中引入基于感受野模块（receptive field block，RFB）改进的RFB-s，通过模仿人类视觉感知，增强特征图的有效感受野区域，提高网络特征表达能力及对目标特征的可辨识性。最后，使用自适应空间特征融合（adaptively spatial feature fusion，ASFF）方式以提升PANet对多尺度特征融合的效果。实验结果表明，在PASCAL VOC数据集上，所提算法检测平均精度均值相较于YOLOv5s提高1.71个百分点，达到84.01%，在满足自动驾驶汽车实时性要求的前提下，在一定程度上减少目标检测时的误检及漏检情况，有效提升模型在复杂驾驶场景下的检测性能。
- Abstract:
- When vehicles, pedestrians, bicycles, and other targets are detected in complex road scenes, the existence of multiscale targets and partial occlusions may easily cause missed and false detections. In this paper, a road target detection algorithm is proposed based on improved YOLOv5s, orienting to autonomous driving scenarios. First, depthwise separable convolution is used to replace partial ordinary convolutions to reduce the number of parameters of the model to improve the detection speed. An improved RFB-s based on receptive field block (RFB) is introduced into the feature fusion network to enhance the effective receptive field area of the feature map, improving the network feature expression capability and the recognizability of the target features by imitating human visual perception. Finally, an adaptive spatial feature fusion method is used to enhance the effect of PANet on multiscale feature fusion. The experimental results reveal that, on the PASCAL VOC dataset, compared with YOLOv5s, the mean value of the average detection precision of the proposed algorithm is improved by 1.71%, reaching 84.01%. Under the premise of meeting the real-time requirement of autonomous driving vehicles, this algorithm has reduced false and missed detections in the target detection to a certain extent, effectively improving the detection performance of the model in complex driving scenarios.
- 参考文献/References:
-
[1] GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition. Columbus: ACM, 2014: 580–587.
[2] GIRSHICK R. Fast R-CNN[C]//2015 IEEE International Conference on Computer Vision. Santiago: IEEE, 2015: 1440–1448.
[3] REN Shaoqing, HE Kaiming, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 39(6): 1137–1149.
[4] DAI Jifeng, LI Yi, HE Kaiming, et al. R-FCN: object detection via region-based fully convolutional networks[C]//Proceedings of the 30th International Conference on Neural Information Processing Systems. Barcelona: ACM, 2016: 379–387.
[5] HE Kaiming, GKIOXARI G, DOLLáR P, et al. Mask R-CNN[C]//2017 IEEE International Conference on Computer Vision. Venice: IEEE, 2017: 2980–2988.
[6] LIU Wei, ANGUELOV D, ERHAN D, et al. SSD: single shot MultiBox detector[C]//European Conference on Computer Vision. Cham: Springer, 2016: 21–37.
[7] FU Chengyang, LIU Wei, RANGA A, et al. DSSD: deconvolutional single shot detector[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Hawaii: IEEE, 2017: 2881-2890.
[8] LI Zuoxin, YANG Lu, ZHOU Fuqiang. FSSD: feature fusion single shot multibox detector[EB/OL]. (2017–12–04)[2022–06–20]. http://arxiv.org/abs/1712.00960.
[9] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 779–788.
[10] REDMON J, FARHADI A. YOLO9000: better, faster, stronger[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017: 6517–6525.
[11] REDMON J, FARHADI A. YOLOv3: an incremental improvement[EB/OL]. (2018–04–08)[2022–06–20]. http://arxiv.org/abs/1804.02767.
[12] BOCHKOVSKIY A, WANG C Y, LIAO H Y M. YOLOv4: optimal speed and accuracy of object detection[J]. (2020–04–23)[2022–06–20]. https://arxiv.org/abs/2004.10934.
[13] GLENN J. Ultralytics. YOLOv5[EB/OL]. (2020–06–03)[2021–04–15]. https://github.com/ultralytics/yolov5.
[14] TAN Mingxing, PANG Ruoming, LE Q V. EfficientDet: scalable and efficient object detection[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2020: 10778–10787.
[15] 陈泽, 叶学义, 钱丁炜, 等. 基于改进Faster R-CNN的小尺度行人检测[J]. 计算机工程, 2020, 46(9): 226–232,241
CHEN Ze, YE Xueyi, QIAN Dingwei, et al. Small-scale pedestrian detection based on improved faster R-CNN[J]. Computer engineering, 2020, 46(9): 226–232,241
[16] 郁强, 王宽, 王海. 一种多尺度YOLOv3的道路场景目标检测算法[J]. 江苏大学学报(自然科学版), 2021, 42(6): 628–633,641
YU Qiang, WANG Kuan, WANG Hai. A multi-scale YOLOv3 detection algorithm of road scene object[J]. Journal of Jiangsu University (natural science edition), 2021, 42(6): 628–633,641
[17] REZATOFIGHI H, TSOI N, GWAK J, et al. Generalized intersection over union: a metric and a loss for bounding box regression[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach: IEEE, 2019: 658–666.
[18] CHOLLET F. Xception: deep learning with depthwise separable convolutions[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017: 1800–1807.
[19] LIU Songtao, HUANG Di, WANG Yunhong. Receptive field block net for accurate and fast object detection[C]//Computer Vision – ECCV 2018: 15th European Conference. Munich: ACM, 2018: 404–419.
[20] SZEGEDY C, LIU Wei, JIA Yangqing, et al. Going deeper with convolutions[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition. Boston: IEEE, 2015: 1–9.
[21] LIU Songtao, HUANG Di, WANG Yunhong. Learning spatial fusion for single-shot object detection[EB/OL]. (2019–11–21)[2022–06–20]. http://arxiv.org/abs/1911.09516.
- 相似文献/References:
-
[1]葛园园,许有疆,赵帅,等.自动驾驶场景下小且密集的交通标志检测[J].智能系统学报,2018,13(3):366.[doi:10.11992/tis.201706040]
GE Yuanyuan,XU Youjiang,ZHAO Shuai,et al.Detection of small and dense traffic signs in self-driving scenarios[J].CAAI Transactions on Intelligent Systems,2018,13():366.[doi:10.11992/tis.201706040]
[2]王星,赵海良,王志刚.基于邻域系统的智能车辆最优轨迹规划方法[J].智能系统学报,2019,14(5):1040.[doi:10.11992/tis.201805004]
WANG Xing,ZHAO Hailiang,WANG Zhigang.Optimal trajectory planning method of intelligent vehicles based on neighborhood system[J].CAAI Transactions on Intelligent Systems,2019,14():1040.[doi:10.11992/tis.201805004]
[3]张新钰,邹镇洪,李志伟,等.面向自动驾驶目标检测的深度多模态融合技术[J].智能系统学报,2020,15(4):758.[doi:10.11992/tis.202002010]
ZHANG Xinyu,ZOU Zhenhong,LI Zhiwei,et al.Deep multi-modal fusion in object detection for autonomous driving[J].CAAI Transactions on Intelligent Systems,2020,15():758.[doi:10.11992/tis.202002010]
[4]陆军,李杨,鲁林超.远距离和遮挡下三维目标检测算法研究[J].智能系统学报,2024,19(2):259.[doi:10.11992/tis.202301001]
LU Jun,LI Yang,LU Linchao.Long-distance and occluded 3D target detection algorithm[J].CAAI Transactions on Intelligent Systems,2024,19():259.[doi:10.11992/tis.202301001]
[5]唐友名,孙冠豫,孙贵斌,等.基于城市超车工况的智能车辆避障规划方法研究[J].智能系统学报,2024,19(3):619.[doi:10.11992/tis.202209060]
TANG Youming,SUN Guanyu,SUN Guibin,et al.Autonomous vehicle trajectory planning based on urban overtaking conditions[J].CAAI Transactions on Intelligent Systems,2024,19():619.[doi:10.11992/tis.202209060]
备注/Memo
收稿日期:2022-06-21。
基金项目:中央高校基本科研业务项目(3122022PY17, 3122017003)；天津市科技计划项目(17ZXHLGX00120)
作者简介:胡丹丹，副教授，主要研究方向为机器人环境感知、多传感器数据融合。申请发明专利30余项，发表学术论文20余篇。E-mail：ddhu@cauc.edu.cn;张忠婷，硕士研究生，主要研究方向为无人驾驶车辆环境感知，被评为校级优秀研究生，曾获国家励志奖学金，华北五省（市、自治区）大学生机器人大赛类人机器人竞技体育赛（投篮）竞赛项目一等奖。E-mail：1113276573@qq.com
通讯作者:胡丹丹. E-mail：ddhu@cauc.edu.cn
更新日期/Last Update:
1900-01-01